def test_encoder_graph_dimensions(dataset, mod_attr, graph_attr):
color_seqs, word_seqs, vocab = dataset
mod = ContextualColorDescriber(vocab, hidden_dim=5, max_iter=1)
mod.fit(color_seqs, word_seqs)
mod_attr_val = getattr(mod, mod_attr)
graph_attr_val = getattr(mod.model.encoder.rnn, graph_attr)
assert mod_attr_val == graph_attr_val
def test_parameter_setting(param, expected):
mod = ContextualColorDescriber(MINIMAL_VOCAB)
mod.set_params(**{param: expected})
result = getattr(mod, param)
if param == "embedding":
assert np.array_equal(result, expected)
else:
assert result == expected
def test_predict_functions_honor_device(dataset, func):
color_seqs, word_seqs, vocab = dataset
mod = ContextualColorDescriber(vocab, max_iter=2)
mod.fit(color_seqs, word_seqs)
prediction_func = getattr(mod, func)
with pytest.raises(RuntimeError):
if func == "predict":
prediction_func(color_seqs, device="FAKE_DEVICE")
else:
prediction_func(color_seqs, word_seqs, device="FAKE_DEVICE")
def test_embedding_update_control(dataset, freeze, outcome):
color_seqs, word_seqs, vocab = dataset
embed_dim = 5
embedding = np.ones((len(vocab), embed_dim))
mod = ContextualColorDescriber(vocab,
max_iter=10,
embedding=embedding,
freeze_embedding=freeze)
mod.fit(color_seqs, word_seqs)
graph_emb = mod.model.decoder.embedding.weight.detach().cpu().numpy()
assert np.array_equal(embedding, graph_emb) == outcome
def test_pretrained_embedding(dataset):
color_seqs, word_seqs, vocab = dataset
embed_dim = 5
embedding = np.ones((len(vocab), embed_dim))
mod = ContextualColorDescriber(vocab,
max_iter=1,
embedding=embedding,
freeze_embedding=True)
mod.fit(color_seqs, word_seqs)
graph_emb = mod.model.decoder.embedding.weight.detach().cpu().numpy()
assert np.array_equal(embedding, graph_emb)
def test_predict_restores_device(dataset, func):
color_seqs, word_seqs, vocab = dataset
mod = ContextualColorDescriber(vocab, max_iter=2)
mod.fit(color_seqs, word_seqs)
current_device = mod.device
assert current_device != torch.device("cpu:0")
prediction_func = getattr(mod, func)
if func == "predict":
prediction_func(color_seqs, device="cpu:0")
else:
prediction_func(color_seqs, word_seqs, device="cpu:0")
assert mod.device == current_device
def test_torch_color_describer_save_load(dataset):
color_seqs, word_seqs, vocab = dataset
mod = ContextualColorDescriber(vocab,
embed_dim=10,
hidden_dim=10,
max_iter=100,
embedding=None)
mod.fit(color_seqs, word_seqs)
mod.predict(color_seqs)
with tempfile.NamedTemporaryFile(mode='wb') as f:
name = f.name
mod.to_pickle(name)
mod2 = ContextualColorDescriber.from_pickle(name)
mod2.predict(color_seqs)
mod2.fit(color_seqs, word_seqs)
dev_cols_test = [
represent_color_context(colors) for colors in dev_rawcols_test
]
# At this point, our preprocessing steps are complete, and we can fit a first model.
# ## Initial model
#
# The first model is configured right now to be a small model run for just a few iterations. It should be enough to get traction, but it's unlikely to be a great model. You are free to modify this configuration if you wish; it is here just for demonstration and testing:
# In[ ]:
dev_mod = ContextualColorDescriber(dev_vocab,
embed_dim=10,
hidden_dim=10,
max_iter=5,
batch_size=128)
# In[ ]:
_ = dev_mod.fit(dev_cols_train, dev_seqs_train)
# As discussed in [colors_overview.ipynb](colors_overview.ipynb), our primary metric is `listener_accuracy`:
# In[ ]:
dev_mod.listener_accuracy(dev_cols_test, dev_seqs_test)
# We can also see the model's predicted sequences given color context inputs:
def test_build_dataset(dataset):
color_seqs, word_seqs, vocab = dataset
mod = ContextualColorDescriber(vocab)
dataset = mod.build_dataset(color_seqs, word_seqs)
result = next(iter(dataset))
assert len(result) == 3
def test_cross_validation_sklearn(dataset):
color_seqs, word_seqs, vocab = dataset
param_grid = {'hidden_dim': [10, 20]}
mod = ContextualColorDescriber(vocab, max_iter=5)
xval = cross_validate(mod, color_seqs, word_seqs, cv=2)
def test_hyperparameter_selection(dataset):
color_seqs, word_seqs, vocab = dataset
param_grid = {'hidden_dim': [10, 20]}
mod = ContextualColorDescriber(vocab, max_iter=5)
xval = RandomizedSearchCV(mod, param_grid, cv=2)
xval.fit(color_seqs, word_seqs)
def test_predict_proba(dataset):
color_seqs, word_seqs, vocab = dataset
mod = ContextualColorDescriber(vocab, max_iter=1)
mod.fit(color_seqs, word_seqs)
probs = mod.predict_proba(color_seqs, word_seqs)
assert all(np.round(t.sum(), 6) == 1.0 for seq in probs for t in seq)
dev_cols_test = [represent_color_context(colors) for colors in dev_rawcols_test]
# At this point, our preprocessing steps are complete, and we can fit a first model.
# ## Initial model
#
# The first model is configured right now to be a small model run for just a few iterations. It should be enough to get traction, but it's unlikely to be a great model. You are free to modify this configuration if you wish; it is here just for demonstration and testing:
# In[18]:
dev_mod = ContextualColorDescriber(
dev_vocab,
embed_dim=10,
hidden_dim=10,
max_iter=5,
batch_size=128)
# In[19]:
#_ = dev_mod.fit(dev_cols_train, dev_seqs_train)
# We can also see the model's predicted sequences given color context inputs:
# In[20]:
from colors import ColorsCorpusReader
import os
import pandas as pd
from sklearn.model_selection import train_test_split
import torch
from torch_color_describer import (ContextualColorDescriber,
create_example_dataset)
import utils
from utils import START_SYMBOL, END_SYMBOL, UNK_SYMBOL
tiny_contexts, tiny_words, tiny_vocab = create_example_dataset(group_size=3,
vec_dim=2)
toy_mod = ContextualColorDescriber(
tiny_vocab,
embedding=None, # Option to supply a pretrained matrix as an `np.array`.
embed_dim=10,
hidden_dim=20,
max_iter=100,
eta=0.01,
optimizer=torch.optim.Adam,
batch_size=128,
l2_strength=0.0,
warm_start=False,
device=None)
_ = toy_mod.fit(tiny_contexts, tiny_words)
metric = toy_mod.listener_accuracy(tiny_contexts, tiny_words)
print("listener_accuracy:", metric)
# In[29]:
toy_color_seqs_train, toy_color_seqs_test, toy_word_seqs_train, toy_word_seqs_test = train_test_split(
toy_color_seqs, toy_word_seqs)
# Here we expose all of the available parameters with their default values:
# In[30]:
toy_mod = ContextualColorDescriber(
toy_vocab,
embedding=None, # Option to supply a pretrained matrix as an `np.array`.
embed_dim=10,
hidden_dim=10,
max_iter=100,
eta=0.01,
optimizer=torch.optim.Adam,
batch_size=128,
l2_strength=0.0,
warm_start=False,
device=None)
# In[31]:
_ = toy_mod.fit(toy_color_seqs_train, toy_word_seqs_train)
# ### Predicting sequences
# The `predict` method takes a list of color contexts as input and returns model descriptions:
# In[32]: